Repository navigation
perf(parser): accelerate line comment scanning - #28878
Conversation
Merging this PR will improve performance by 3.42%1 warning
Performance Changes
Tip Curious why performance improved? Comment Comparing Footnotes
|
Merge activity
|
Codex Review Summary: β CompletedThis comment shows the latest Codex review activity on this pull request.
βΉοΈ About Codex in GitHubYour team has set up Codex to review pull requests in this repo. Reviews are triggered when you
Codex reacts with π while any review is running, comments if it has suggestions, and reacts with π once all reviews finish with no findings. |
Use `memchr3` to scan line comments, retaining the scalar scanner for Unicode candidates. On M3 Max, production parse time falls 1.3β1.4% (geometric mean across 39 Oxc/SWC fixtures). Native 128-byte/1 KB comment cases fall 49%/79%; dense Unicode is ~1.3% slower. Validated conformance, allocations, lint, Rust/Node suites, and 566,440 differential cases. AI-assisted.
5ffd472 to
296b12a
Compare
### π Bug Fixes - de6ac92 isolated-declarations: Preserve object accessor mutability (#29039) (camc314) - 3729ce1 transformer/typescript: Merge exported enums across namespace blocks (#28999) (camc314) - 7243594 minifier: Preserve throwing ArrayBuffer and URI calls (#29012) (camc314) - 67ac6be isolated-declarations: Evaluate enum template literal initializers (#28997) (camc314) - f05b30e isolated-declarations: Infer auto-accessor initializer types (#29038) (camc314) - 030d410 isolated-declarations: Preserve computed ambient enum members (#29040) (camc314) - 6303df1 jsdoc: Preserve param tags after unmatched description delimiters (#28923) (Vishal Punjabi) - 99e67d3 transformer: Preserve static private brands with shadowed class names (#28894) (Dunqing) - 595f3c2 minifier: Preserve left-hand side effects in instanceof checks (#29011) (camc314) - a791a5b cfg: Resolve continues through stacked loop labels (#29016) (camc314) - 9e3b3d1 ecmascript: Reject signs after nondecimal prefixes (#29034) (camc314) - 43c5273 ecmascript: Round nondecimal strings to numbers once (#29026) (camc314) - b71234f cfg: Skip labeled contexts for unlabeled jumps (#29001) (camc314) - fd51bff jsdoc: Handle every line terminators as newline (#28902) (leaysgur) - c00dc61 codegen: Keep one source map mapping per generated position (#28844) (YT) - 52f3e3d jsdoc: Do not treat `@word` containing `/` as tag (#28834) (leaysgur) - 0b48483 minifier: Preserve UTF-16 boundaries in empty-search replaceAll folding (#28807) (huuya) - 1e57a4e codegen: Print infinite numeric literals as `1e400` instead of `Infinity` (#28812) (Dominic Elm) - 28a44b7 transformer/class-properties: Preserve private accessor assignment receivers (#28790) (Dunqing) - ea76490 transformer/class-properties: Preserve private method values in assignments (#28788) (Dunqing) - 8432de6 transformer/class-properties: Apply a compound assignment to a static private field once (#28606) (YT) - 3dbf140 parser: Treat escaped contextual keywords as names, not modifiers (#28748) (Dominic Elm) - e3b1bb4 transformer/class-properties: Keep the value of a private accessor assignment (#28716) (YT) - ae7474e minifier: Keep array concat chain reads before later side effects (#28261) (Dunqing) - ec327aa semantic: Keep block type bindings separate from hoisted vars (#28601) (YT) - 3fd78d7 codegen: Preserve names in duplicate source mappings (#26051) (camc314) ### β‘ Performance - 7761a39 codegen: Remove redundant sourcemap buffer shrinks (#28893) (Dunqing) - bee25de regular_expression: Check only the latest group per name in duplicate prescan (#29083) (Yuji Sugiura) - 4da1752 codegen: Use SWAR scanning and shared column tables (#28939) (Dunqing) - cc84126 mangler: Inline candidate checks during slot assignment (#28881) (Dunqing) - 6611cf9 parser: Specialize quoted string value access (#28965) (Boshen) - 5038ba6 parser: Avoid repeated await-using lookahead (#28960) (Boshen) - ecc807b parser: Skip discarded arrow diagnostics (#28940) (Boshen) - c880368 parser: Fast-path absent TypeScript type lists (#28959) (Boshen) - dee6cf4 parser: Skip repeated optional call lookahead (#28876) (Boshen) - a524e17 parser: Reuse successful parenthesized arrow heads (#28862) (Boshen) - 296b12a parser: Accelerate line comment scanning (#28878) (Boshen) - c4fc21e parser: Accelerate escaped string scanning (#28907) (Boshen) - f7726b6 parser: Accelerate regex literal scanning (#28911) (Boshen) - a8aff38 parser: Scan whitespace in runs (#28953) (Boshen) - b1c137c parser: Avoid lookahead for TypeScript keyword types (#28954) (Boshen) - 6a8a2a7 parser: Fast-path plain identifiers (#28955) (Boshen) - 00deb87 parser: Initialize declaration lists with their first item (#28956) (Boshen) - 71dd877 parser: Specialize binary operator associativity (#28957) (Boshen) - 732a8e4 codegen: Skip repeated word scans for sourcemap tails (#28869) (Dunqing) - 6972243 codegen: Accelerate sourcemap line scanning (#28400) (codspeed) - 027eb7e parser: Combine async arrow lookahead (#28431) (Boshen) - 7b933d5 lexer: Directly dispatch common punctuation (#28792) (Boshen) - 45d86c3 lexer: Combine per-token annotation resets (#28492) (Boshen) - d497fbf lexer: Match keywords with byte patterns (#28494) (Boshen) ### π Documentation - 844c28c syntax: Correct crate capabilities and module records (#29075) (camc314) - 8d30d9b traverse: Clarify ancestor access during traversal (#29074) (camc314) - a74b7cd ecmascript: Clarify test locations (#29072) (camc314) - 7acaa4e mangler: Correct naming strategy and pipeline order (#29073) (camc314) - ed28291 ast: Clarify which AST nodes use arena allocation (#29070) (camc314) - d5832ab oxc: Correct README APIs and feature requirements (#29068) (camc314) - 261c4a2 allocator: Clarify arena storage and memory reclamation (#29066) (camc314) - 6448a54 napi/parser: Correct parser options and usage guidance (#29065) (camc314) - 157fc5a napi/minify: Correct minifier capabilities and caveats (#29064) (camc314) - 6989796 parser: Update TS-ESTree import differences (#29061) (camc314)
Use
memchr3to scan line comments, retaining the scalar scanner for Unicode candidates.On M3 Max, production parse time falls 1.3β1.4% (geometric mean across 39 Oxc/SWC fixtures). Native 128-byte/1 KB comment cases fall 49%/79%; dense Unicode is ~1.3% slower.
Validated conformance, allocations, lint, Rust/Node suites, and 566,440 differential cases.
AI-assisted.